b=20918 improve log warning

[fs/lustre-release.git] / lustre / tests / recovery-mds-scale.sh

diff --git a/lustre/tests/recovery-mds-scale.sh b/lustre/tests/recovery-mds-scale.sh
index 7440ed5..0ab9999 100644 (file)
--- a/lustre/tests/recovery-mds-scale.sh
+++ b/lustre/tests/recovery-mds-scale.sh
@@ -14,9 +14,13 @@ CLEANUP=${CLEANUP:-""}
 init_test_env $@
 
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
 
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-mds-scale}
+TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
 DEBUGLOG=$TESTSUITELOG.debug
+
+cleanup_logs
+
 exec 2>$DEBUGLOG
 echo "--- env ---" >&2
 env >&2
@@ -28,9 +32,9 @@ set -x
 
 [ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
 [ $CLIENTCOUNT -ge 3 ] || \
-    { skip "$0 Need two or more clients, have $CLIENTCOUNT" && exit 0; }
+    { skip "$0 Need two or more remote clients, have $CLIENTCOUNT" && exit 0; }
 
-END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY}/end_run_file}
+END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
 
 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
@@ -47,25 +51,30 @@ rm -rf $DIR/[df][0-9]*
 # -- remove hostname from clients list
 zconf_umount $(hostname) $MOUNT
 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
-NODES_TO_USE=$(exclude_item_from_list $NODES_TO_USE $(hostname))
+NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $(hostname))
 
 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
 
-MDTS=""
-for ((i=1; i<=$MDSCOUNT; i++)) do
-    MDTS="$MDTS mds$i"
-done
-MDTS=$(comma_list $MDTS)
-
-OSTS=""
-for ((i=1; i<=$OSTCOUNT; i++)) do
-    OSTS="$OSTS ost$i"
-done
-OSTS=$(comma_list $OSTS)
+MDTS=$(get_facets MDS)
+OSTS=$(get_facets OST)
 
 ERRORS_OK=""    # No application failures should occur during this test.
 FLAVOR=${FLAVOR:-"MDS"}
 
+if [ "$FLAVOR" == "MDS" ]; then
+    SERVERS=$MDTS
+else
+    SERVERS=$OSTS
+fi
+ 
+if [ "$SLOW" = "no" ]; then
+    DURATION=${DURATION:-$((60 * 30))}
+    SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
+else
+    DURATION=${DURATION:-$((60 * 60 * 24))}
+    SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
+fi
+
 rm -f $END_RUN_FILE
 
 vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
@@ -101,7 +110,7 @@ summary_and_cleanup () {
     # actually failed though.  the first node in the END_RUN_NODE is
     # the one we are really interested in.
         if [ -n "$END_RUN_NODE" ]; then
-            var=${END_RUN_NODE}_load
+            var=$(client_var_name $END_RUN_NODE)_load
             echo "Client load failed on node $END_RUN_NODE" 
             echo
             echo "client $END_RUN_NODE load stdout and debug files :
@@ -140,6 +149,15 @@ Status: $result: rc=$rc"
         sleep 5
         kill -9 $CLIENT_LOAD_PIDS || true
     fi
+    if [ $rc -ne 0 ]; then
+        # we are interested in only on failed clients and servers
+        local failedclients=$(cat $END_RUN_FILE | grep -v $0)
+        # FIXME: need ostfailover-s nodes also for FLAVOR=OST
+        local product=$(gather_logs $(comma_list $(osts_nodes) \
+                               $(mdts_nodes) $mdsfailover_HOST $failedclients))
+        echo logs files $product
+    fi
+
     [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
 
     exit $rc
@@ -152,7 +170,6 @@ log "-----============= $0 starting =============-----"
 
 trap summary_and_cleanup EXIT INT
 
-DURATION=${DURATION:-$((60*60*24))}
 ELAPSED=0
 NUM_FAILOVERS=0
 
@@ -176,16 +193,6 @@ fi
 START_TS=$(date +%s)
 CURRENT_TS=$START_TS
 
-if [ "$FLAVOR" == "MDS" ]; then
-    SERVER_FAILOVER_PERIOD=$MDS_FAILOVER_PERIOD
-    SERVERS=$MDTS
-else
-    SERVER_FAILOVER_PERIOD=$OSS_FAILOVER_PERIOD
-    SERVERS=$OSTS
-fi
-
-SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
-
 MINSLEEP=${MINSLEEP:-120}
 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3}  # bug17839 comment 62
 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
@@ -213,7 +220,20 @@ while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
         exit 4
     fi
 
-    log "Starting failover on $SERVERNODE"
+    log "Wait $SERVERFACET recovery complete before doing next failover ...."
+    if [[ $NUM_FAILOVERS != 0 ]]; then
+        if ! wait_recovery_complete $SERVERFACET ; then
+            echo "$SERVERFACET recovery is not completed!"
+            exit 7
+        fi
+    fi
+
+    log "Checking clients are in FULL state before doing next failover"
+    if ! wait_clients_import_state $NODES_TO_USE $SERVERFACET FULL; then
+        echo "Clients import not FULL, please consider to increase SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD !"
+        
+    fi
+    log "Starting failover on $SERVERFACET"
 
     facet_failover "$SERVERFACET" || exit 1
 
@@ -242,11 +262,19 @@ while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
     if [ $sleep -lt $MINSLEEP ]; then
         reqfail=$((reqfail +1))
         log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
-Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
+Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
+This iteration, the load was only applied for sleep=$sleep seconds.
+Probably the hardware is taking excessively long to boot.
+Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
         [ $reqfail -gt $REQFAIL ] && exit 6 
     fi  
 
     log "$SERVERFACET has failed over ${!var} times, and counting..."
+
+    if [ $((ELAPSED + sleep)) -gt $DURATION ]; then
+         break
+    fi
+
     if [ $sleep -gt 0 ]; then 
         echo "sleeping $sleep seconds ... "
         sleep $sleep